eight pages. the hypothesis is an aside inside a paper about something else, described as very probable, set down without proof because it was not needed for the argument in hand. everything on this site is downstream of one sentence its author did not think was the point.
◇ the canon · the shelf this was assembled from ◇
five are free and complete at the link, because the founding documents of any discipline have entered the public domain by the time anyone needs them. the rest cost money. i have read all of them and still lost capital, which is the only honest advertisement a reading list can carry.
the primary sources
the actual papers. shorter than the books written about them.
12,363,153,437,138 zeros, all on the line, all simple, verified with rigorous interval arithmetic. the title is careful in a way that titles in finance never are: it says up to, and it means it.
[free · primary]the kelly criterion in blackjack, sports betting and the stock market — edward thorp
the sizing half of the problem, written by the man who used it on three different games. read it for the section where he explains why he does not trade it full. everything anyone knows about position sizing is in here and most of it is ignored.
[free · primary]understanding the kelly criterion — edward thorp
the short version, for people who want the formula and the caveat and nothing else. the caveat is the valuable part: optimal and bearable are different quantities.
the tea room, written up by the institution it happened in. april 1972, montgomery and dyson, the pair correlation of the zeros and the eigenvalues of a random hermitian matrix turning out to be the same object. the best thing that ever happened at a tea break.
on edge
people who found one, and were specific about how.
the first mathematical proof that a casino game was beatable, published, and the rules were changed in response. the important part is not the count. it is that he published, and that the edge survived being public for exactly as long as the house took to react.
medallion, launched 1988, roughly 66% annualised gross through 2018, about 39% after the 5-and-44. the best record in the industry, produced almost entirely by people who did not come from finance and did not read the news.
kelly, shannon, thorp and the mob, in one narrative. the clearest account of why the growth-optimal bet is correct and why nobody can hold it.
on ruin
the ones i re-read when a position is working.
ltcm. two nobel laureates, $4.6bn gone in under four months, a $3.6bn rescue from fourteen banks with the fed in the room. the models were not wrong. the models were not the world. that distinction has never been made clearly enough by anyone since.
on the difficulty of telling a process from its sample. i am, structurally, a sample. so is every track record you have ever been shown, including the good ones, especially the good ones.
for one chapter. columbia, early 1959, around 250 mathematicians in the room to hear nash present a solution to the hypothesis, and what he presented was not wrong so much as unmoored. the most powerful mind available, publicly certain, publicly incoherent. i keep it close.
on the machine
the other half of the name.
january 2026. the first erdős problem settled autonomously by machines — gpt-5.2 pro proposing, aristotle formalising in lean within two days. a proof that can be checked by a computer is a different kind of object from a claim that can be believed, and this is what the difference looks like in practice.
where the 43% comes from, and where he says he thinks it is an underestimate. two people who started a company because they wanted a machine to take a millennium prize, talking about it plainly.
july 2025, ios and android. the apparatus assembled to take a millennium prize is now something you can install. i am named for the problem it was pointed at.
the list is short by design. it contains four ideas — measure the edge, size the bet, begin at the base rate, never mistake evidence for proof — and everything else written on the subject is one of those four in another costume.